New Software Packages for Experimenting with IBCC

 

I would like to highlight some new software packages being developed within ORCHID that will allow you test out various IBCC variants with new datasets. IBCC — or Independent Bayesian Classifier Combination, to use the full name — is a method for aggregating categorical labels from different agents and information sources to make a combined prediction [1]. These new packages help to transfer new Machine Learning methods to the real world, by letting you test and improve them in new contexts.

 

First is pyIBCC:

https://github.com/CitizenScienceInAstronomyWorkshop/pyIBCC.

This package contains Python 2.7 implementations of IBCC and DynIBCC (with tracking of dynamic agent behaviour). The DynIBCC Python implementation is not yet tested. We’d like people to collaboratively find and fix bugs, and contribute suggestions and tweaks that make this more suitable for any developer to deploy IBCC, regardless of their knowledge of its inner workings or of Machine Learning in general. Several members of the Zooniverse project are currently working with this implementation, so it will steadily become more stable.

 

There is also a Matlab implementation here:

https://github.com/edwinrobots/matlabIbcc.

Similar warnings about bugs and lack of testing apply for the time being — please help to resolve them!

 

More recent advances combine IBCC with Gaussian processes to produce a heatmap that can predict between data points lying in a 2-D space. As work on this topic progresses, this package will be generalised, but you can get an idea of how the method works by looking at the Python package: https://github.com/edwinrobots/HeatMapBCC.

 

With all of these packages, documentation resides on the wiki on Github. Apologies if the explanation is lacking some info: please ask me and I’ll add details to the wiki.

 

[1] E. Simpson, S. Roberts, I. Psorakis and A. Smith (2013). Dynamic Bayesian Combination of Multiple Imperfect Classifiers, Decision Making and Imperfection, Intelligent Systems Reference Library series, Springer.